Methodology / SCI
How SCI scoring works.
The Skill Credibility Index is a transparent, event-sourced score that measures demonstrated skill from five evidence types. This page explains the full scoring model.
Design principles
SCI was designed around three constraints. First, every score must be reproducible from raw events. Given the same evidence history, the same SCI must result. Second, no single evidence type can dominate. A user who only takes assessments cannot reach Authority tier without session evidence. Third, recent activity matters more than historical activity because skills decay.
These constraints prevent gaming. You cannot buy your way to a high SCI. You cannot accumulate endorsements from inactive accounts. You cannot submit a proof once and coast indefinitely.
The scoring formula
SCI is a weighted composite of five evidence signals, each normalized to a 0-1000 scale before weighting. The final score is the weighted sum, clamped to [0, 1000].
Scoring formula
Each component score is itself a function of count, quality, and recency. A session rated 5/5 contributes more than one rated 3/5. Recent sessions contribute more than sessions from six months ago.
Recency decay
Evidence events lose weight over time. We apply an exponential decay function with a half-life of 90 days. An event from today has full weight. An event from 90 days ago has 50% weight. An event from 180 days ago has 25% weight.
This means SCI reflects your current skill level, not your historical peak. If you stop practicing, your score naturally declines. If you stay active, your score stays strong. This is intentional: skill credibility should be a living metric.
SCI measures what you can do now, not what you could do once.
Versioning and reproducibility
The scoring algorithm is versioned. When we change weights or decay parameters, we increment the version. Old scores are preserved under their original version. We can replay any score from raw events and detect drift between computed and stored values.
Daily reconciliation jobs compare stored projections against recomputed values. If drift exceeds a configurable threshold, the system automatically triggers a backfill. This guarantees that every SCI score you see is accurate within the current algorithm version.
Evidence types and weights
Sessions
Completed peer sessions where you teach or learn a skill. Each session generates a rating, duration confirmation, and completion event. Session evidence is the strongest signal because it proves live, interactive competence.
Assessments
Structured skill tests administered through the platform. Assessments verify knowledge depth on specific topics. Results are time-stamped and versioned so we can track improvement over time.
Endorsements
Peer endorsements weighted by the endorser's own SCI score. An endorsement from someone at Authority tier carries more weight than one from a Newcomer. This creates a credibility network, not a popularity contest.
External Proofs
Verifiable artifacts from outside Lemma: GitHub repositories, published papers, certifications, portfolio links. Each proof is validated for authenticity and linked to the claimed skill.
Portfolio
Work samples uploaded directly to your Skill Passport. Portfolio items provide tangible evidence of applied skill. They are reviewed for relevance but not subjectively graded.
Tier thresholds
Newcomer
Proven
Verified
Expert
Authority
Just joined. No verified evidence yet.
Initial evidence recorded. Early pattern forming.
Consistent evidence across multiple sessions.
Strong, sustained track record. Trusted by peers.
Exceptional depth. Recognized across the platform.
SCI scoring FAQ
Yes. The formula is public and versioned. SCI is a weighted sum of five normalized evidence signals: sessions (40%), assessments (20%), endorsements (15%), external proofs (15%), and portfolio (10%). Each signal uses recency decay with a 90-day half-life.
Start proving what you know.
Early access is rolling out for individuals and teams. No credit card, no PDFs — just the things you made, made visible.